Effects of Language Modeling and its Personalization on Touchscreen Typing Performance

Abstract

Modern smartphones correct typing errors and learn userspecific words (such as proper names). Both techniques are useful, yet little has been published about their technical specifics and concrete benefits.One reason is that typing accuracy is difficult to measure empirically on a large scale. We describe a closed-loop, smart touch keyboard (STK) evaluation system that we have implemented to solve this problem. It includes a principled typing simulator for generating human-like noisy touch input, a simple-yet-effective decoder for reconstructing typed words from such spatial data, a large web-scale background language model (LM), and a method for incorporating LM personalization. Using the Enron email corpus as a personalization test set, we show for the first time at this scale that a combined spatial/language model reduces word error rate from a pre-model baseline of 38.4% down to 5.7%, and that LM personalization can improve this further to 4.6%.

Ubiquitous Text Interaction

“…Interfaces also need to provide users with powerful, fluid, and task-appropriate text interaction. Prominent amongst these are supporting editing and correcting [1,9,25,18], formatting of text [7], personalization [8], and context-awareness [10,11].…”

A Monte Carlo simulation approach for quantitatively evaluating keyboard layouts for gesture input

“…In addition, language model personalization could be incorporated and compared to the overall effect of keyboard layout on the error rate. Fowler et al showed that including language model personalization can dramatically reduce the word error rate for touch typing [Fowler et al (2015)]. …”

“…The methodology that we propose allows for the direct evaluation of gesture reconstruction error rates, or any other desired metric, by simulating realistic user interactions with a keyboard. This is similar to the approach used by Fowler et al when they simulated noisy tap typing input to estimate the effect of language model personalization on word error rate [Fowler et al (2015)]. …”

On Suggesting Phrases vs. Predicting Words for Mobile Text Composition

“…Word prediction has been shown to help people enter text more quickly and accurately, showing keystroke savings of up to 45% in both mobile keyboards [10] and accessibility domains [17,34,33]. Despite its utility and near-ubiquity on mobile devices, word prediction has received relatively little research attention in mobile text entry.…”

“…Typical mobile word-prediction systems use small bigram models (e.g., [10]), though one commercial predictive keyboard has implemented neural language models that are able to utilize earlier context information 1 . With the benefit of longer context, statistical language modeling has been shown to be capable of accurately predicting phrases [26,7].…”